5 分鐘看懂如何用卷積神經網路 CNN 判讀肺炎 CXR 影像

In Coding, Healthcare By Sigrid On

川流不息的病人、處理不完的 Complaints、堆積待打的病歷，此時電話那頭護理師提醒：「醫師可以幫我看一下 X 光片嗎？」

無論在醫院見實習當住院醫師，還是去診所做健檢業務或一般看診，胸腔 X 光影像（CXR）就跟每天都會刷牙洗臉吃飯如影隨形（咦 🤔？），尤其從 2019 年至今已經舉世造成極大影響的 COVID 就是從肺炎起家，各科英雄好漢懷疑病人有肺部感染時，除了兔寶寶戳鼻子採檢和血單、Culture 之外，就來張 CXR 起手式吧！

川流不息的病人、處理不完的 Complaints、堆積待打的病歷，此時電話那頭護理師提醒：「醫師可以幫我看一下 X 光片嗎？」

攝影師：cottonbro，連結：Pexels

當然好 🖖！不過，如果有個小幫手可以先協助我做點功課，把五花八門的可能性縮減到較窄的範圍，好讓我點開 PACS 時已經有了初步的答案，是不是聽起來不錯呢？那麼，這次就讓我們來瞧瞧如何用 CNN 來建構肺炎閱片小幫手吧 👀

我們使用的是 Kaggle 肺炎胸腔 X 光資料集，來自於 2018 年發表在 Cell 學術期刊的論文，主要使用的是已經 Label 好的兒童胸部 X 光片，5,232 張影像裡分成 1,349 張正常、2,538 張細菌性肺炎、1,345 張病毒性肺炎的影像，大家可以點進去、下載資料集來玩玩看。

Kaggle 肺炎胸腔 X 光資料集

多元分類資料集的準備

如果大家已經下載了資料集，開始訓練模型前，我們先來看看手邊有哪些資料可以用。它已經很貼心地分好 Train / Validation / Test 三個資料夾，而每個資料夾裡又有「Normal」、「Pneumonia（肺炎）」兩種影像，Pneumonia 裡又根據檔名的註記，分為 Bacteria（細菌）與 Virus（病毒）兩個類別。

看完了資料長相，怎麼準備資料集呢？我們先來設定基本參數，如圖片大小固定成每張都是 200×200 再送給 Model、定義好 3 個類別（正常、細菌性、病毒性，當然也可以只分成正常和肺炎的二元分類），接著把圖片讀進來、以 for Loop 用路徑 Split 的方式指定每張圖片對應的分類。

IMG_SIZE = 200
all_class = ['normal', 'bacteria', 'virus']
class_map = {cls:i for i,cls in enumerate(all_class)} #  'normal':0, 'bacteria': 1, 'virus':2

# read all paths
img_paths_train = glob('chest_xray/train/*/*.jpeg')
img_paths_val = glob('chest_xray/val/*/*.jpeg')
img_paths_test = glob('chest_xray/test/*/*.jpeg')

# Resize img
img_resized = cv2.resize(img, (IMG_SIZE, IMG_SIZE))

def read_data(paths):
    data_count = len(paths)
    x = np.zeros((data_count, IMG_SIZE, IMG_SIZE, 3))
    y = np.zeros((data_count, ))

    for i, path in tqdm(enumerate(paths)):
        # read image
        img = cv2.imread(path)
        img = cv2.resize(img, (IMG_SIZE, IMG_SIZE)) # resize
        img = img / 255. # normalization

        # read class index
        cls = path.split(os.sep)[-2]
        # for pneumonia class
        if cls == 'pneumonia':
            # get filename
            filename = path.split(os.sep)[-1]
            # get pneumonia class
            cls_pneumonia = filename.split('_')[1] 
            cls_idx = class_map[cls_pneumonia]
        # for normal class
        else:
            cls_idx = class_map[cls]
        x[i] = img
        y[i] = cls_idx
    return x, y
    
x_train, y_train = read_data(img_paths_train)
x_val, y_val = read_data(img_paths_val)

不過我們現在 y 裡裝的只有最開始 class_map 的 0, 1, 2，要做 one-hot encoding 轉為長度 = 3（分幾類決定了長度）的向量，用 keras 的 utils 中to_categorical 這個 Function。

# one-hot encoding
y_train = utils.to_categorical(y_train, num_classes=len(class_map))
y_val = utils.to_categorical(y_val, num_classes=len(class_map))
y_test = utils.to_categorical(y_test, num_classes=len(class_map))

建構模型

準備好資料以後，就來建模吧！要特別注意，餵給模型的 Data Input Size，形狀必須跟我們原本資料集 x_train, x_val, x_test 一樣才行。以下面這個模型為例，前兩層的卷積我們使用 64 個 Filters、Kernel Size 是 3×3、Activation Function 用 Relu，接著加上 1 層 MaxPooling…重複再用 2 層分別有 128、256、512 個 Filters 的卷積加上 1 層 MaxPooling，最後把 Feature Maps 的特徵向量攤平 Flatten、接上 1 層 Dense Layer，分成 3 類、以 softmax 輸出模型結果。

inputs = layers.Input(shape=(IMG_SIZE, IMG_SIZE, 3))
x = layers.Conv2D(filters=64, kernel_size=3, activation='relu')(inputs)
x = layers.Conv2D(filters=64, kernel_size=3, activation='relu')(x)
x = layers.MaxPool2D(2)(x)
x = layers.Conv2D(filters=128, kernel_size=3, activation='relu')(x)
x = layers.Conv2D(filters=128, kernel_size=3, activation='relu')(x)
x = layers.MaxPool2D(2)(x)
x = layers.Conv2D(filters=256, kernel_size=3, activation='relu')(x)
x = layers.Conv2D(filters=256, kernel_size=3, activation='relu')(x)
x = layers.MaxPool2D(2)(x)
x = layers.Conv2D(filters=512, kernel_size=3, activation='relu')(x)
x = layers.Conv2D(filters=512, kernel_size=3, activation='relu')(x)
x = layers.MaxPool2D(2)(x)
x = layers.Flatten()(x)
x = layers.Dense(64, activation='relu')(x)
prediction = layers.Dense(3, activation='softmax')(x)
model = models.Model(inputs=inputs, outputs=prediction)

訓練模型

建模後，我們設定讓模型計算 Loss、優化超參數、衡量結果的指標，用分類的損失函數 categorical_crossentropy、Optimizer 用預設的 SGD（Stochastic Gradient Descent）、計算模型的 Accuracy，再設定 Batch Size、Epochs 的數量。

model.compile(loss=tf.keras.losses.categorical_crossentropy,
              optimizer=tf.keras.optimizers.SGD(),
              metrics=['accuracy'])
              
logs = model.fit(x_train, y_train,
                batch_size=32,
                epochs=50,
                verbose=1,
                validation_data=(x_val, y_val),
                 )

跑完之後，我們找出訓練歷程中 Validation Loss 最低是多少、對應的 Accuracy 有多高，再把 Training（藍色線）、Validation 隨著 50 個 Epochs 的 Iterations，Accuracy 和 Loss 的變化是多少；通常 Training 的 Loss 會比較低。

min_loss_epoch = np.argmin(history['val_loss'])
print('val loss ', history['val_loss'][min_loss_epoch])
print('val acc', history['val_accuracy'][min_loss_epoch])

plt.plot(history['accuracy'])
plt.plot(history['val_accuracy'])
plt.legend(['accuracy', 'val_accuracy'])
plt.title('accuracy')

plt.plot(history['loss'])
plt.plot(history['val_loss'])
plt.legend(['loss', 'val_loss'])
plt.title('loss')
Accuracy 隨著 Iteration 的變化
Loss 隨著 Iteration 的變化

衡量模型的其他 Metrics

sklearn 有提供分類任務裡常用的 classification_report、Confusion Matrix，我們用模型沒看過的資料（x_test），讓模型預測每一筆資料的機率值，再透過 argmax 看哪一個類別有最大的機率（y_pred），去跟 Ground Truth（y_true）來評估成果如何。像我們總體的 Accuracy 只有 0.6，第 0 類表現最好（0.73）、第 1 類表現最糟（0.41）。

from sklearn.metrics import classification_report, confusion_matrix

y_true = np.argmax(y_test, axis=-1)
prediction = model.predict(x_test)
y_pred = np.argmax(prediction, axis=-1)

# classification_report
print(classification_report(y_true, y_pred))
這次小幫手看他沒看過的 CXR 結果的分類報告

希望這次的分享大家覺得有意思，下次再讓我們繼續看看 AI 遇上臨床工作人員，有什麼令人怦然心動的火花吧 🌹

#CC

2,874 thoughts on “5 分鐘看懂如何用卷積神經網路 CNN 判讀肺炎 CXR 影像”

  296. Keep a look out for an online bingo bonus – Many of the best online bingo sites offer attractive bingo bonuses. No deposit bonuses, deposit bonuses, free tickets, BOGOF bonuses, and more are all worth considering. PLEASE NOTE THE LEGAL AGE TO PLAY BINGO IS 18+. you can experience all of our over 900+ online slots and casino games on all iOS and Android mobile devices through your browser, meaning you can take your favourite slots and casino games wherever you like. Considered one of the best live dealer casinos on the market, the LeoVegas Live Casino is a great destination for both low and high-rollers who enjoy numerous games and attractive tailor-made bonuses. The platform is available on a top-notch mobile app offering 24 7 authentic casino action. You can email the site owner to let them know you were blocked. Please include what you were doing when this page came up and the Cloudflare Ray ID found at the bottom of this page.
    https://preniumdirectory.com/listings12764112/pragmatic-slots-news
    Mobile casinos tend to have fewer games than their desktop counterparts. However, regardless of whether you’re using a desktop computer or mobile device, we expect the best gambling operators to offer over 100 games. Add to this in-app features, such as search filters, the ability to favorite games, and you’ve got a top-rated platform. Basically, if your mobile experience is worse or doesn’t exceed what’s available via a casino’s website, it doesn’t receive a high score. Launching the same day as Parx in July 2019, Hollywood Casino was joint first to market in Pennsylvania. Hollywood’s online casino and mobile app are powered by IGT, an international gaming brand with decades of experience in the industry. The casino’s mobile offering is easy to get started with, and laid out cleanly. It came out of the gate stocked with popular slots and even a progressive, along with blackjack, baccarat, and video poker.

    Reply

  430. The community feature is designed to boost player engagement and retention, as it rewards both the jackpot winner and the last 1,000 players of the jackpot game. Jackpot Inferno features 5 reels and a 50-payout grid that is commonly present in many penny slots. The lowest bet in the game is 0.30 coins and the highest bet is 3.00. What really makes Jackpot Inferno an awesome slot machine game is all of the frequently triggered bonus rounds. There is a Jackpot Jump Progressive bonus and the Free Spin bonus. When you win in the main game of Jackpot 6000, players have a 1 in 2 shot at doubling their winnings in the feature Heads or Tails. Win in Heads or Tails and you can continue doubling your profits or take your winnings and move on to the Super Meter game. Click the Heads or Tails button to let all of your winnings ride or only use a part of them by clicking on the transfer button. Each time you hit transfer, 20 credits are saved. Players who do not want to play Heads or Tails can click the collect button to cash in all winnings and go back to the main Jackpot 6000 game.
    https://datos.olacefs.com/user/erpitthankni1972
    Dubbed by IGT as the “payout beast,” Siberian Storm lives up to its name by offering 720 ways to win in its MultiWay Xtra feature, along with free spins, wilds, and scatters. The slot incorporates a Siberian tiger theme, complete with icy landscapes and snow-capped forests. While with free slot machines you can hone your skills and perfect your strategy, there is one big drawback: you can’t win any money! Real money slot machines can sometimes offer life-changing sums of money to players, and even the smaller winnings can intensify the excitement. If you’re unsure whether you’d like to try real money slot machines or stick with playing free casino slot games, we’ve detailed the benefits of both in the table below: Dubbed by IGT as the “payout beast,” Siberian Storm lives up to its name by offering 720 ways to win in its MultiWay Xtra feature, along with free spins, wilds, and scatters. The slot incorporates a Siberian tiger theme, complete with icy landscapes and snow-capped forests.

    Reply